Re: Concurrency, Exceptions and Poison

[ianeast <ianeast@xxxxxxxxxxxxxxx>]

I suspect many of these problems result from the idea of a process being 
the source of an exception, which is surely just another event. The only 
difference is the need for interruption.

I've never liked the idea of "throwing" an exception. If a process can be 
the source of an event to which it must react then it is necessarily 
non-deterministic. Hoare (p. 178) talks about these problems right from 
the start of dealing with non-determinism.

An alternative is NOT to allow a process to source any event to which it 
must react. Events such as division by zero could instead be communicated 
by a distinct "watchman" process. The only problem with this is that 
interruption may not _immediately_ follow, allowing damage to perhaps be 
done.

Ian


Dr. Ian Robert East         School of Computing and Mathematical Sciences
ireast@xxxxxxxxxxxxx                            Oxford Brookes University
(44) 1865 483635                                           Oxford OX3 0BP

Consultation hours for 2001/2002 Term 1
To be decided